Sample-specific fragment GC bias and other changes

[mikelove]

The main change here is to allow fragment GC bias, as explained in a new extras argument frag_GC_bias to simulate_experiment().

Other changes include:

• allow 0 reads output
• strand_specific as an extras argument to avoid reverse complementing
• fraglen and fragsd can be either length 1 vectors or length equal to the number of samples
• write out the pre-fragment-GC-downsampled counts matrix to outdir

[JMF47]

Thanks Mike!